ai-memory-core

ai-memory-core

Persistent memory infrastructure for AI coding agents. A 68-tool MCP server that gives OpenCode, Claude Code, and Cursor agents memory that persists across sessions — error tracking, decision logging, skill routing, task orchestration, multi-modal AI generation, and behavioral verification.

Built on mem0 for cloud-backed cross-session memory and agent-memory-mcp for local project-specific conventions.

🌟 Features

🧠 Memory & Persistence

🤖 Multi-Modal AI Generation

🧩 Agent Infrastructure

🚀 For Non-Technical Users

You only need Docker. Nothing else. No Node.js, no Python, no npm, no pip.

Windows

Send your friend this link — it downloads the file, then they double-click it:

https://github.com/ohmpatel3877/ai-memory-core/releases/download/v1.0.0/ONE-CLICK.cmd

This is a GitHub Release link, so the browser will download it (not display it). They double-click the downloaded file and it runs.

Mac / Linux

Open Terminal, paste this one line, press Enter:

curl -fsSL https://raw.githubusercontent.com/ohmpatel3877/ai-memory-core/main/docker/setup-opencode-container.sh | bash

What they get

A running container with the 68-tool MCP server. Connect it to OpenCode by adding to opencode.json:

{
  "mcpServers": {
    "opencode-container-server": {
      "command": "docker",
      "args": ["exec", "-i", "opencode-server", "python3", "/app/scripts/tools-mcp-server.py"]
    }
  }
}

📦 Developer Install

Prerequisites

• Node.js 18+
• Python 3.10+
• A mem0 API key (free tier available)

Windows (PowerShell 7+)

git clone https://github.com/ohmpatel3877/ai-memory-core.git
cd ai-memory-core
pwsh plugin-tools/install-ai-memory-core.ps1

macOS / Linux

git clone https://github.com/ohmpatel3877/ai-memory-core.git
cd ai-memory-core
bash plugin-tools/install-ai-memory-core.sh

Docker

docker compose up -d

The installer will:

1. ✅ Validate Node.js + Python versions
2. ✅ Install npm dependencies
3. ✅ Prompt for MEM0_API_KEY (or read from .env)
4. ✅ Register the MCP server in opencode.json
5. ✅ Link skills to ~/.config/opencode/skills/
6. ✅ Run verification checks

🛠 MCP Server — 68 Tools

The MCP server runs on the standard JSON-RPC protocol over stdio. OpenCode and Claude Code auto-discover it from opencode.json.

Core Infrastructure (13 tools)

Coder (7 tools)

coder_analyze_code, coder_generate_framework, coder_debug, coder_review, coder_explain, coder_convert, coder_architecture

Supports: Python, JavaScript, TypeScript, Rust, Go, Java, C#, C++, Swift, Kotlin, Ruby, PHP

Sensory — Web & File I/O (12 tools)

sensory_browse, sensory_screenshot, sensory_interact, sensory_extract_pdf, sensory_extract_html, sensory_extract_image, sensory_scrape, sensory_extract_article, sensory_api_request, sensory_fetch_rss, sensory_read_file, sensory_search

Headless Firefox via Playwright. DuckDuckGo search — no API key required.

DevOps — Infrastructure (7 tools)

devops_container_debug, devops_permissions_analyze, devops_compose_generator, devops_samba_config, devops_mergerfs_setup, devops_dockerfile_analyze, devops_network_troubleshoot

Audio (7 tools)

audio_analyze_file, audio_waveform, audio_frequency_analysis, audio_music_theory, audio_speech_analysis, audio_convert_guide, audio_generate_tone

Game Dev (7 tools)

gamedev_design_analyze, gamedev_scaffold_project, gamedev_mechanics_guide, gamedev_monetization, gamedev_optimization, gamedev_compare_engines, gamedev_level_design

Art (4 tools)

art_generate_svg, art_generate_theme, art_extract_palette, art_design_concept

Literature (4 tools)

lit_analyze_text, lit_extract_concepts, lit_generate_study_guide, lit_analyze_philosophy

Memory Search (5 tools)

memory_search, memory_synthesize, memory_add, memory_consolidate, memory_status

Verifier (2 tools)

verifier_status, verifier_renudge

🧩 Skills

Skill Router Triggers

The router auto-loads skills based on keywords in your task description:

📐 Architecture

┌─────────────────────────────────────────────────────┐
│                    OpenCode / Claude Code             │
├─────────────────────────────────────────────────────┤
│                    MCP Protocol (stdio)              │
├─────────────────────────────────────────────────────┤
│                  tools-mcp-server.py                  │
│  ┌──────┐ ┌──────┐ ┌──────┐ ┌──────┐ ┌──────────┐  │
│  │ Core │ │Coder │ │Audio │ │ Art  │ │  Sensory  │  │
│  │13 tls│ │ 7 tls│ │ 7 tls│ │4 tls │ │  12 tls  │  │
│  └──────┘ └──────┘ └──────┘ └──────┘ └──────────┘  │
│  ┌──────┐ ┌────────┐ ┌──────┐ ┌────────┐ ┌──────┐  │
│  │DevOps│ │Game Dev│ │ Lit  │ │Memory  │ │Verif │  │
│  │ 7 tls│ │ 7 tls  │ │4 tls │ │Search  │ │2 tls │  │
│  └──────┘ └────────┘ └──────┘ └──┬─────┘ └──────┘  │
├───────────────────────────────────┼─────────────────┤
│        mem0 (Cloud-Backed)        │  Agent-Memory-MCP│
│   ┌─────────────────────────┐     │  ┌─────────────┐ │
│   │ Cross-session memory    │     │  │ Project-local│ │
│   │ User preferences        │     │  │ Conventions  │ │
│   │ Anti-patterns           │     │  │ ADR history  │ │
│   │ Resolved bugs           │     │  │ State files  │ │
│   └─────────────────────────┘     │  └─────────────┘ │
└───────────────────────────────────┴───────────────────┘

Memory Categories

ai-memory-core uses 16 custom memory categories with configurable retention:

🚀 Usage Examples

Track an Error Across Sessions

# Agent A encounters an error
→ xtrace_log_error("ModuleNotFoundError: No module named 'xyz'")

# Agent B (next session) searches before debugging
→ xtrace_search("ModuleNotFoundError")
  ← "Already logged. Root cause: missing pip install. Fix: pip install xyz"

Record an Architecture Decision

→ dtrace_add(
    title="Choose PostgreSQL over MySQL",
    context="Need JSONB, array columns, and CTEs for the analytics module",
    decision="PostgreSQL 16",
    alternatives=["MySQL 8", "SQLite"],
    consequences="Slightly higher ops overhead, better query capability"
  )

Auto-Route a Task to Skills

# User says: "debug the Samba container networking issue"
# Skill Router auto-loads: troubleshooting-master, devops_network_troubleshoot
# No manual skill selection needed

Decompose a Complex Task

# Task Orchestrator scores complexity, splits into parallel workstreams:
# [Agent 1: Frontend] React component redesign
# [Agent 2: Backend] API endpoint rewrite
# [Agent 3: DB] Schema migration
# Merge results when all complete

📋 Scripts

All CLI scripts live in scripts/:

🔧 Customization

Memory Categories

Edit .mem0.md to adjust retention policies or add categories:

architecture_decisions: null       # Keep forever
anti_patterns: 365                 # 1 year
session_state: 90                  # 3 months

Skill Router

Edit skills/skill-router.json to add trigger rules:

{
  "triggers": ["kubernetes", "k8s", "helm", "pod"],
  "skills": ["kubernetes-operations-k8s-manifest-generator"],
  "priority": 10
}

MCP Server Config

Edit package.json → mcpServers or opencode.json to change server args:

"ai-memory-core": {
  "command": "python",
  "args": ["scripts/tools-mcp-server.py"]
}

🔗 Related Projects

📄 License

MIT © Ohm Patel

_{Built with mem0, OpenCode, and too much coffee.}